OpenAI discloses it blocked extraction attacks stealing reasoning content, but they remain reproducible on Azure
Understand the story
On September 30, 2026, OpenAI said via official channels that it had disrupted a coordinated attempt to extract the inference capabilities of its protected models, and that it would strengthen its defenses against adversarial distillation. On October 1, The Decoder reported more details: from July 24 to 25, over 16,000 requests were made by more than 4,000 users, involving more than 15,000 linked accounts; OpenAI said these were all attempted extractions, that it had shut down the related activity entirely on July 28, and that it had linked the core group to people associated with Moonshot AI (the developer of the Kimi model). The report also noted that the same technique could still be used to steal inference content on Azure. The current state of the event: OpenAI has disclosed and shut down the related activity, while the Azure vulnerability remains reproducible; no further developments have been reported.
AI overview of reporting · Updated 17 days ago
Reporting timeline
Explore different aspects of the story through its reporting.
- The DecoderOpenAI says it stopped a campaign to steal its models' reasoning, but the trick still worked on Azure
OpenAI disclosed a distillation theft campaign targeting its models' reasoning content: from July 24-25, more than 16,000 requests were made by over 4,000 users, involving more than 15,000 linked accounts. OpenAI said it shut them all down on July 28 and linked the core group to people associated with Moonshot AI (developer of the Kimi model), while noting these were attempted extractions.
- OpenAI NewsDisrupting a coordinated model-distillation campaign
OpenAI said it has disrupted a coordinated campaign attempting to extract the reasoning capabilities of its protected models, and will strengthen its defenses against adversarial distillation.